TWI801717B - A physical image generation method and device, device, non-transitory computer-readable storage medium and computer program product - Google Patents

Abstract

The invention discloses a method, device, equipment, non-transitory computer-readable storage medium and computer program product for generating a physical image. The method includes: performing eigendecomposition on a first physical image to obtain the first physical image The first reflected image and the first illuminated image; performing at least one transformation on at least one of the first reflected image and the first illuminated image; according to the transformed at least one image, the first reflected image and the first illuminated image to generate at least one second physical image.

Description

A physical image generation method and device, device, non-transitory computer-readable storage medium and computer program product

The invention belongs to the field of computer vision, and in particular relates to a method, device, equipment, non-transitory computer-readable storage medium and computer program product for generating an object image.

Image recognition is a technology that uses computers to process, analyze and understand images to identify targets and objects in various patterns. For example, face recognition and identity verification. Image recognition models need to use a large number of physical pictures for training.

For an image recognition model, the more pictures of real objects, the richer the scenes covered by the training data, and the more accurate the recognition of real objects. However, the manual collection of physical pictures has great limitations. It is impossible to obtain real pictures in some scenes by adjusting the shooting conditions, especially the slight changes in the shooting conditions. The physical pictures in some scenes are missing, and the training materials are incomplete, which in turn causes the physical recognition model to fail to accurately recognize the real objects in these missing scenes.

Therefore, in the prior art, manual collection of physical images results in the lack of physical images in some scenes, and the problem of incomplete training data needs to be solved urgently.

Embodiments of the present invention provide a physical image generation method, device, equipment, non-transitory computer-readable storage medium, and computer program products, which solve the problem of missing physical images in some scenes caused by manual collection of physical images in the prior art, and training The problem of incomplete information.

In the first aspect, an embodiment of the present invention provides a method for generating an object image, including: performing eigendecomposition on the first object image, and acquiring a first reflection image and a first illumination image of the first object image; At least one of the first reflected image and the first illuminated image is transformed at least once; and at least one first reflected image is generated based on the transformed at least one image, the first reflected image, and the first illuminated image Two physical images.

Optionally, performing at least one transformation on at least one of the first reflection image and the first illumination image includes: according to the M first pixel value transformation rules in the preset illumination transformation algorithm, transforming the The pixel values in the first illuminated image undergo M different transformations to obtain different M transformed second illuminated images; wherein, each transformation in the M different transformations is the same as the M transformed second One second irradiation image in the irradiation image is uniquely corresponding; M is a positive integer; and, according to the transformed at least one image, the first reflection image and the first irradiation image, at least one second irradiation image is generated. The real object image includes: generating M second real object images different from the first real object image according to the first reflected image and the M transformed second irradiation images.

Optionally, performing at least one transformation on at least one of the first reflection image and the first illumination image includes: according to the N second pixel value transformation rules in the preset reflection transformation algorithm, transforming the The pixel values in the first reflection image are transformed differently N times to obtain different N transformed second reflection images; wherein, each transformation in the N times of different transformations is the same as the first transformation after the N transformations A second reflection image in the two reflection images is uniquely corresponding; N is a positive integer; and the Generating at least one second object image according to the transformed at least one image, the first reflected image and the first illuminated image, comprising: according to the first illuminated image and the N transformed second Reflecting the image, generating N second physical images different from the first real image.

Optionally, performing at least one transformation on at least one of the first reflection image and the first illumination image includes: according to the P third pixel value transformation rules in the preset illumination transformation algorithm, transforming the The pixel values in the first illuminated image are transformed differently for P times to obtain different P transformed third illuminated images; wherein, each transformation in the P different transforms is the same as the P transformed first A third irradiation image in the three irradiation images is uniquely corresponding; P is a positive integer; according to the Q fourth pixel value conversion rule in the preset reflection transformation algorithm, Q is performed on the pixel value in the first reflection image different transformations to obtain different Q transformed third reflected images; wherein, each transformation in the Q different transformed transformations is related to a third illuminated image in the Q transformed third illuminated images The image is uniquely corresponding; Q is a positive integer; and generating at least one second physical image according to the transformed at least one image, the first reflected image and the first illuminated image includes: according to the P transformations The final third irradiation image and the Q transformed third reflection images generate P*Q second physical images different from the first real image.

Optionally, after generating at least one second physical image, it also includes: inputting the at least one second physical image as training data into the image recognition model; according to each second physical image in the training data The parameters of the image recognition model are updated with the output result after inputting the second real object image into the image recognition model.

In the above method, by performing eigendecomposition on the first real object image, the first reflected image and the first illuminated image of the first real object image are obtained, and then the first reflected image and the first illuminated image are At least one image in the image is transformed at least once, so it is possible to obtain the transformed At least one of the illuminated image and the reflection image, and then the transformed at least one image, the first reflection image and the first illumination image are combined with each other, so that at least one Physical images, and so on, carry out the above steps for each physical image collected manually, which can greatly improve the physical image, make up for the lack of manual collection of physical images in some scenes, and achieve the effect of supplementing the training data .

In a second aspect, an embodiment of the present invention provides a device for generating an image of a real object, including: an acquisition module configured to perform eigendecomposition on a first image of an actual object, and acquire the first reflection image and the second reflected image of the first object image. An illuminated image; a processing module, configured to perform at least one transformation on at least one of the first reflected image and the first illuminated image; The reflected image and the first illuminated image generate at least one second physical image.

Optionally, the processing module is specifically used to: perform M different transformations on the pixel values in the first illumination image according to the M first pixel value transformation rules in the preset illumination transformation algorithm, and obtain different M transformed second illumination images; wherein, each transformation in the M different transformations is uniquely corresponding to one of the M transformed second illumination images; M is a positive integer ; Generate M second real object images different from the first real object image according to the first reflected image and the M transformed second illuminated images.

Optionally, the processing module is specifically used to: perform N times of different transformations on the pixel values in the first reflection image according to the N second pixel value transformation rules in the preset reflection transformation algorithm, and obtain different The N transformed second reflection images; wherein, each transformation in the N different transformations is uniquely corresponding to one of the N transformed second reflection images; N is positive An integer; according to the first illuminated image and the N transformed second reflection images, generate N second real object images different from the first real object image.

Optionally, the processing module is specifically used to: perform P different transformations on the pixel values in the first illumination image according to the P third pixel value transformation rules in the preset illumination transformation algorithm, and obtain different The P transformed third illumination images; wherein, each transformation in the P different transformations is uniquely corresponding to one of the P transformed third illumination images; P is positive Integer; according to the Q fourth pixel value transformation rules in the preset reflection transformation algorithm, perform Q different transformations on the pixel values in the first reflection image to obtain different Q transformations of the third reflection image ; Wherein, each transformation in the Q times of different transformations is uniquely corresponding to a third irradiation image in the Q transformed third irradiation images; Q is a positive integer; according to the P transformed third The irradiated image and the Q transformed third reflection images generate P*Q second physical images different from the first real image.

Optionally, the processing module is also used to: input the at least one second physical image as training data into the image recognition model; according to each second physical image in the training data, the first 2. An output result after inputting the real object image into the image recognition model, and updating the parameters of the image recognition model.

In a third aspect, an embodiment of the present invention provides a physical image generation device, including: at least one processor; and a memory connected in communication with the at least one processor; wherein, the memory stores information that can be processed by the at least one processor Instructions executed by a processor, the instructions are executed by the at least one processor, so that the at least one processor can execute the method for generating an image of a physical object described in the first aspect above.

In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute Carry out the physical image generation method described in the first aspect above.

In a fifth aspect, an embodiment of the present invention provides a computer program product, the computer program product includes a computing program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by the computer, The computer is made to execute the method for generating an image of a physical object described in the first aspect above.

201-203: Steps

401-404: Steps

501-506: steps

601: Get the module

602: processing module

700: Physical image generation equipment

701: Transceiver

702: Processor

703: memory

704: bus bar system

FIG. 1 is an overall flow chart of a method for generating an object image provided in Embodiment 1 of the present invention;

FIG. 2 is a flow chart of the steps of a method for generating an object image provided in Embodiment 1 of the present invention;

Fig. 3 is a schematic diagram corresponding to eigendecomposition of a physical image generation method provided by Embodiment 1 of the present invention;

FIG. 4 is a flow chart of the steps of a method for generating an object image provided in Embodiment 2 of the present invention;

FIG. 5 is a flow chart of the steps of a method for generating an object image provided by Embodiment 3 of the present invention;

FIG. 6 is a schematic structural diagram of a physical image generating device applied to Embodiments 1, 2 and 3 of the present invention;

FIG. 7 is a schematic structural diagram of an object image generating device applied to Embodiments 1, 2 and 3 of the present invention.

In order to facilitate the review committee to understand the technical features, content and advantages of the present invention and the effects it can achieve, the present invention is hereby combined with the accompanying drawings and appendices, and is described in detail in the form of an embodiment. The description is as follows, and the purpose of the diagrams used therein is only for illustration and auxiliary instructions, and may not be the true proportion and precise configuration of the present invention after implementation, so the proportion and configuration relationship of the attached diagrams should not be interpreted , Limit the scope of application of the present invention on the actual implementation, together describe first.

In describing the present invention, it is to be understood that the terms "center", "lateral", "upper", "lower", "left", "right", "top", "bottom", "inner", " The orientations or positional relationships indicated in the drawings are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, Specific orientation configurations and operations, therefore, are not to be construed as limitations on the invention.

Image recognition is widely used in the field of computer vision, such as various face verification systems, by identifying the face, obtaining the permission of the identity, and then operating it. The tool that realizes this function of image recognition is an image recognition model.

Before realizing the image recognition function, the image recognition model needs to use a large number of physical pictures for training. For an image recognition model, the more pictures of real objects, the richer the scenes covered by the training data, and the more accurate the recognition of real objects. However, the manual collection of physical pictures has great limitations. It is impossible to obtain real pictures in some scenes by adjusting the shooting conditions, especially the slight changes in the shooting conditions. The physical pictures in some scenes are missing, and the training materials are incomplete, which in turn causes the physical recognition model to fail to accurately recognize the real objects in these missing scenes.

Therefore, the embodiment of the present invention proposes a method for generating more physical images based on a small number of manually collected physical images. As shown in FIG. 1 , it is an overall flowchart of a method for generating an object image provided in an embodiment of the present invention. It should be noted that Figure 1 only uses a manually collected physical image as an example to illustrate the process, and the manually collected physical images are also displayed in multiple scenarios. Collected.

Given a real object image, use the intrinsic decomposition algorithm to decompose the real object image into a shading image (Shading image) and a reflection image (Reflectance image). Among them, the physical image is the initial image collected manually; the illuminated image is the image that reflects the illumination of the original image; the reflected image refers to the image that can remain unchanged under changing illumination conditions, reflecting the original object. Image textures, materials, etc. Among them, the physical image, the irradiated image and the reflected image are all composed of multiple pixels, each pixel has a pixel value, and the combination of each pixel forms an image and produces a visual effect. Each pixel point has a corresponding pixel value in the real object image, the irradiation image and the reflection image, and each pixel value in the real object image, the irradiation image and the reflection image corresponds to each other.

Taking a manually collected physical image as an example, after obtaining an irradiated image and a reflected image after eigendecomposition, the reflected image and the irradiated image are transformed several times. Obtain a reflected image or illuminated image whose pixel values are not exactly the same as those of the original reflected image or illuminated image, and use these modified illuminated and reflected images to generate a large number of images different from the original image set. Real image. Among them, the brightness of the artificially collected physical image is generally affected by the ambient light, while the material information of the physical object itself has nothing to do with the lighting conditions. Therefore, the embodiment of the present invention transforms the reflection image into different lighting conditions to obtain multiple transformed reflection images under different lighting conditions; post-irradiation image. It should be noted that, the above-mentioned specific ways of transforming the reflected image or the illuminated image are realized by transforming the pixel values of the pixels in the reflected image or the illuminated image through a preset algorithm.

The above method for generating an object image will be further described in detail below in conjunction with FIG. 2 .

As shown in FIG. 2 , it is a flow chart of steps of a method for generating an object image provided in an embodiment of the present invention.

Step 201: Perform eigendecomposition on the first object image, and acquire a first reflection image and a first illumination image of the first object image.

In step 201, the relationship between the first physical image (I), the first reflected image (R), and the first illuminated image (S) can be expressed by a formula:

I(x,y)=S(x,y)*R(x,y) (1)

Where (x, y) is the pixel coordinate of the pixel in the image. According to this formula, the reflection map (R) and the irradiation map (S) are numerically solved, and the schematic diagram is shown in Figure 3, which is a schematic diagram of the eigendecomposition corresponding to a first physical image generation method provided in the embodiment of the present invention . The first row is the first real object image, the middle row is the first reflected image, and the last row is the first illuminated image. It should be noted that, since only the pixel values of the pixel coordinates in the first real image I are known, the pixel coordinates correspond to more than one set of pixel values of the first illuminated image and the first reflected image Therefore, in the intrinsic decomposition process, the decomposed pixel values of the first illuminated image and the pixel values of the first reflected image are a set of randomly selected solutions.

Step 202: Perform at least one transformation on at least one of the first reflected image and the first illuminated image.

Step 203: Generate at least one second physical object image according to the transformed at least one image, the first reflected image and the first illuminated image.

In step 202, three situations are included: in the first situation, at least one transformation takes M transformations as an example, and the pixel values in the first illuminated image are processed according to the M first pixel value transformation rules in the preset illumination transformation algorithm. M different transformations, get Take different M transformed second illumination images; wherein, each transformation in M different transformations is uniquely corresponding to one of the M transformed second illumination images; M is a positive integer ; In the second case, according to the N second pixel value transformation rules in the preset reflection transformation algorithm, N times of different transformations are performed on the pixel values in the first reflection image, and different N transformed first pixel values are obtained. Two reflection images; wherein, each transformation in the N different transformations is uniquely corresponding to one of the N transformed second reflection images; N is a positive integer; in the third case, According to the P third pixel value conversion rules in the preset illumination transformation algorithm, P different transformations are performed on the pixel values in the first illumination image to obtain different P transformed third illumination images; wherein , each transformation in the P times of different transformations uniquely corresponds to one of the P transformed third illumination images; P is a positive integer.

In addition, according to the Q fourth pixel value transformation rules in the preset reflection transformation algorithm, Q times of different transformations are performed on the pixel values in the first reflection image to obtain Q different transformed third reflection images ; Wherein, each transformation in the Q times of different transformations uniquely corresponds to a third illumination image among the Q transformed third illumination images; Q is a positive integer.

In the above three cases, take the first case as an example, in this case, the default illumination transformation algorithm is packaged in an image processing software, such as openCV. When the image processing software invokes the light condition transformation algorithm, there are multiple first pixel value transformation rules corresponding to the light condition transformation, that is, one light condition corresponds to one first pixel value transformation rule. By transforming the pixel values of the first illumination image according to a first pixel value transformation rule, a transformed illumination image under corresponding lighting conditions can be obtained. In the second and third cases, pixel value conversion rules are preset according to lighting conditions or textures, and reflection images and illuminated images with changed lighting conditions or textures are obtained by changing pixel values, so details will not be repeated here.

In step 203, respectively corresponding to the situation in step 202, including the following three situations: the first situation, after performing the first situation in step 202, according to the first reflection image and the M transformed second irradiation image, generating M second real object images different from the first real object image; in the second case, after performing the second case in step 202, according to the first illuminated image and the N transformed The second reflected image of the first real image is generated to generate N second real images different from the first real image; the third case, after performing the third case in step 202, according to the P transformed third The irradiated image and the Q transformed third reflection images generate P*Q second physical images different from the first real image.

To sum up, there are three situations in which the physical images are generated from steps 202 to 203, which are expressed by formulas as follows: In the first situation, keep the first reflection image (R) obtained in the intrinsic decomposition stage unchanged, and for the first The irradiation image (S) is modified differently to obtain M times different second irradiation images (AS ⁽ⁱ⁾ ), and then the real object image (AI ⁽ⁱ⁾ ) is calculated using the formula (1): AI ^{(i )} (x,y)=AS ⁽ⁱ⁾ (x,y)*AR(x,y),i=1,...,M(2) In this case, by modifying the lighting conditions of the first illuminated image , which generates a physical image set:

A=[AI ⁽¹⁾ ,AI ⁽²⁾ ,...,AI ^(M) ]; in the second case, the first illumination image (S) obtained in the intrinsic decomposition stage remains unchanged, and the first reflection The image (R) is modified N times to obtain different reflection images (BR _j ), and then the generated image (BI _j ) is calculated using the formula (1): BI _j (x,y)=BS(x, y)*BR _j (x,y), j=1,...,N; (3) In this case, by modifying the texture of the first reflection image, a physical image set is generated

B=[BI ₁ ,BI ₂ ,...,BI _N ]: In the third case, Q times of different modifications are performed on the first reflected image (R) to obtain a different third reflected image (CR _j ) , keep each third reflected image (CR _j ) unchanged, and make P different modifications to the first illuminated image (S) to get a different third illuminated image (CS _j ), then use the formula (1 ) to calculate the generated image (CI _j ):

得到生成資料集

get generated data set

After step 203, another optional implementation manner is to input the at least one second physical image as training data into the image recognition model; according to each second physical image in the training data, the first 2. An output result after inputting the real object image into the image recognition model, and updating the parameters of the image recognition model. Through the generated second physical image, the amount of training data is greatly increased, which can make the image recognition model more accurate.

As shown in FIG. 4 , it is a flow chart of steps of a method for generating an object image provided by Embodiment 2 of the present invention. Embodiment 2 of the present invention is a method for generating a multi-illumination face image based on eigendecomposition. Illumination change is the most critical factor affecting the performance of face recognition, and the degree to which the problem is solved is related to the success or failure of the practical process of face recognition. In order to improve the robustness of the face recognition model to illumination, one of the most direct methods is to add face images under different illumination conditions to the training data. The specific steps are as follows: Before step 401, collect a physical image obtained by manual photography Image set E, for example, E contains 100,000 face images; Step 401: Perform eigendecomposition on each object image in object image set E.

In step 401, for example, k=1,2...100000, perform eigendecomposition on each picture EI _k in the physical image set E, and obtain the corresponding reflection image (ER _k ), illumination map Like (ES _k ).

Step 402: keep the reflected image (ER _k ) unchanged, and perform n different modifications to the illuminated image (ES _k ) according to a preset lighting condition modification algorithm. Wherein, n is an integer greater than 1.

In step 402, each illumination image ES _k obtains a transformed illumination image set

Step 403: Generating a physical object image set according to the transformed irradiation image set and reflection image.

Then use the following formula to generate a collection of physical images

；i=1,2...n。

;i=1,2...n.

Step 404: Determine whether there are images of objects in the data set E that have not been subjected to steps 402 and 403.

If yes, go to step 402; otherwise, use the physical image set generated by each physical image in E as the final generated training data set E _g . Taking 100,000 pictures in the physical image as an example, E _g =[E ₁ ,...,E ₁₀₀₀₀₀ ], which contains a total of 1 million pictures; use the data set [E,E _g ] to train the physical recognition model, Obtain an object recognition model that is more robust to lighting conditions.

FIG. 5 is a flow chart of steps of a method for generating an object image provided by Embodiment 3 of the present invention. Embodiment 3 of the present invention is a method for generating training data for image segmentation based on eigendecomposition. The purpose of image segmentation is to divide the image into regions with different characteristics and extract the target of interest. These features can be pixels, colors, textures, etc., and the extraction targets can be single or multiple regions. The specific steps are as follows: Before step 501, collect a physical image set F obtained by manual shooting. For example, F contains 1000 landscape images; Sign decomposition.

In step 501, for example, m=1,2...1000, perform eigendecomposition on each picture FI _m in the physical image set F, and obtain the corresponding reflection image (FR _m ), illumination map Like (FS _m ).

Step 502: keep the reflected image (FR _m ) unchanged, and perform t different modifications to the illuminated image (FS _m ) according to the preset lighting condition modification algorithm. It should be noted that the preset lighting condition transformation algorithm includes multiple pixel value transformation rules, and each pixel value transformation rule corresponds to a transformed reflection image.

In step 502, the transformed irradiation image set is obtained

Step 503: Generating a real object image set according to the transformed irradiation image set

Step 503 utilizes the following formula to transform:

，i=1,...,t。

, i=1,...,t.

Step 504: keep the illuminated image (FS _m ) unchanged, and perform r different modifications to the reflected image (FR _m ) according to a preset texture modification algorithm. It should be noted that the preset texture transformation algorithm includes multiple pixel value transformation rules, and each pixel value transformation rule corresponds to a transformed reflection image.

In step 504, the transformed reflection image set is obtained

[FR _m,1 ,...,FR _m,r ].

Step 505: Generate a physical image set according to the transformed reflection image set

F _m ' =[FI _m,1 ,...,FI _m,r ].

In step 505, the following formula is utilized for transformation:

FI _m,j (x,y)=FS _m (x,y)*FR _m,j (x,y),j=1,...,r.

Step 506: Determine whether there are images of real objects in the data set F that have not undergone step 502 and step 504. If yes, go to step 502; otherwise, use the physical image set generated by each physical image in E as the final training data set F _h =[F ₁ ,F ₁ ' ...,F _t ,F _r ' ]. For example, when t=r=10, F _h contains a total of 20,000 physical images. Using the data set [F, F _h ] to train the image segmentation model, and using the data set with richer lighting conditions, colors, and material conditions for model training will greatly improve its accuracy.

In the above method, by performing eigendecomposition on the first real object image, the first reflected image and the first illuminated image of the first real object image are obtained, and then the first reflected image and the first illuminated image are At least one image in the image is transformed at least once, so at least one transformed irradiation image and reflected image can be obtained, and then the transformed at least one image, the first reflected image and the first illuminated image The images are combined with each other, so that at least one physical image can be generated from the initial first physical image, and so on, the above steps are carried out for each physical image collected manually, which can greatly improve the physical image and make up for part of the scene. The lack of manual collection of physical images can achieve the effect of supplementing the training data.

A method of generating a large amount of training data based on a small amount of manually collected physical pictures in Embodiments 1, 2 and 3 of the present invention, by performing eigendecomposition on the manually collected physical pictures Obtain the illumination map and reflection map, modify the illumination map and reflection map according to the change of illumination conditions or texture, so as to generate a physical picture containing more abundant types of illumination and texture, and expand the image recognition model. The training data makes the image recognition model more accurate and robust to the recognition of objects in different scenarios.

The embodiment of the present invention greatly reduces manpower input; through the above method, a large number of physical images can be generated as training data, thereby greatly reducing the cost of data accumulation, and collecting a large amount of training data in a relatively short period of time. In addition, a large amount of training data can be customized according to specific application scenarios; this method can generate a training data set containing more abundant lighting and texture types on the basis of the original physical image set; the generated data can be trained to A model that is more robust to lighting effects and more general for different scenes can improve the performance of computer vision fields such as face or object detection and recognition, and image segmentation.

As shown in FIG. 6 , it is a schematic structural diagram of an object image generating device applied in Embodiments 1, 2 and 3 of the present invention.

An embodiment of the present invention provides an object image generating device, including: an acquisition module 601, configured to perform eigendecomposition on a first object image, and obtain a first reflection image and a first illumination map of the first object image image; a processing module 602, configured to perform at least one transformation on at least one of the first reflected image and the first illuminated image; and the first illuminated image to generate at least one second object image.

Optionally, the processing module 602 is specifically configured to: perform M different transformations on the pixel values in the first illumination image according to the M first pixel value transformation rules in the preset illumination transformation algorithm, and obtain Different M transformed second illuminated images; wherein, in the M different transforms Each transformation is uniquely corresponding to one of the M transformed second irradiation images; M is a positive integer; according to the first reflection image and the M transformed second irradiation images , generate M second physical images different from the first real image.

Optionally, the processing module 602 is specifically configured to: perform N times of different transformations on the pixel values in the first reflection image according to the N second pixel value transformation rules in the preset reflection transformation algorithm, and obtain Different N transformed second reflection images; wherein, each transformation in the N different transformations is uniquely corresponding to one of the N transformed second reflection images; N is A positive integer; according to the first illuminated image and the N transformed second reflection images, generate N second real object images different from the first real object image.

Optionally, the processing module 602 is specifically configured to perform P different transformations on the pixel values in the first illumination image according to the P third pixel value transformation rules in the preset illumination transformation algorithm, and obtain Different P transformed third irradiation images; wherein, each transformation in the P different transformations is uniquely corresponding to a third irradiation image in the P transformed third irradiation images; P is Positive integer; according to the Q fourth pixel value transformation rules in the preset reflection transformation algorithm, perform Q times of different transformations on the pixel values in the first reflection image, and obtain different Q transformations of the third reflection image image; wherein, each transformation in the Q times of different transformations is uniquely corresponding to a third illumination image in the Q transformed third illumination images; Q is a positive integer; according to the P transformed first The three illuminated images and the Q transformed third reflected images generate P*Q second physical images different from the first real image.

Optionally, the processing module 602 is also configured to: input the at least one second physical image as training data into the image recognition model; according to each second physical image in the training data, and the The output result after the second physical image is input to the image recognition model, and the figure is updated Like the parameters of the recognition model.

Based on the same technical concept, an embodiment of the present invention provides a real object image generating device. at least one processor; and, a memory connected in communication with the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one The processor can execute the method for generating a real object image in the above-mentioned embodiments.

Taking a processor as an example, FIG. 7 shows the structure of a physical image generation device provided by an embodiment of the present invention. The physical image generation device 700 includes: a transceiver 701, a processor 702, a memory 703, and a bus system 704; Among them, the memory 703 is used to store programs. Specifically, the program may include program code, and the program code includes computer operation instructions. The memory 703 may be a random access memory (random access memory, RAM for short), or a non-volatile memory (non-volatile memory), such as at least one disk memory. Only one memory is shown in the figure, of course, multiple memories can also be set as required. The memory 703 may also be a memory in the processor 702 .

The memory 703 stores the following elements, executable modules or data structures, or their subsets, or their extended sets: Operation instructions: including various operation instructions, used to implement various operations.

Operating system: includes various system programs for implementing various basic businesses and processing hardware-based tasks.

The method for generating a real object image in the embodiment of the present invention described above may be applied to the processor 702 , or implemented by the processor 702 . The processor 702 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above-mentioned method for generating an image of a real object can be processed by The integrated logic circuit of the hardware in the processor 702 or instructions in the form of software are implemented. The above-mentioned processor 702 may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. Various methods, steps and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like. The steps of the methods disclosed in connection with the embodiments of the present invention may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as random access memory, flash memory, read-only memory, programmable read-only memory or electrically readable and writable programmable memory, register. The storage medium is located in the memory 703, and the processor 702 reads the information in the memory 703, and performs the following steps in combination with its hardware: the transceiver 701 is used to perform eigendecomposition on the first object image, and obtain the first the first reflected image and the first illuminated image of the real object image; the processor 702, configured to perform at least one transformation on at least one of the first reflected image and the first illuminated image; and for At least one second physical object image is generated based on the transformed at least one image, the first reflected image and the first illuminated image.

Preferably, the processor 702 is specifically configured to: perform M different transformations on the pixel values in the first illumination image according to the M first pixel value transformation rules in the preset illumination transformation algorithm, and obtain different M transformed second irradiation images; wherein, each transformation in the M different transformations is uniquely corresponding to a second irradiation image in the M transformed second irradiation images; M is a positive integer; according to The first reflection image and the M transformed second irradiation images generate M second real object images different from the first real object images.

Preferably, the processor 702 is specifically configured to: perform N times of different transformations on the pixel values in the first reflection image according to the N second pixel value transformation rules in the preset reflection transformation algorithm, and obtain different N Transformed second reflection images; wherein, each transformation in the N different transformations is uniquely corresponding to one of the N transformed second reflection images; N is a positive integer; According to the first illuminated image and the N transformed second reflected images, N second real object images different from the first real object images are generated.

Preferably, the processor 702 is specifically configured to perform P different transformations on the pixel values in the first illumination image according to the P third pixel value transformation rules in the preset illumination transformation algorithm, and obtain different P A transformed third irradiation image; wherein, each transformation in the P different transformations is uniquely corresponding to a third irradiation image in the P transformed third irradiation images; P is a positive integer; According to the Q fourth pixel value transformation rules in the preset reflection transformation algorithm, Q times of different transformations are performed on the pixel values in the first reflection image to obtain different Q transformed third reflection images; wherein , each transformation in the Q times of different transformations uniquely corresponds to one of the Q transformed third irradiation images; Q is a positive integer; according to the P transformed third irradiation images image and the Q transformed third reflection images to generate P*Q second physical images different from the first real image.

Preferably, the processor 702 is further configured to: input the at least one image of the second object into the image recognition model as training data; The output result after the image is input to the image recognition model updates the parameters of the image recognition model.

The physical image generation device in the embodiment of the present invention exists in various forms, including but not limited to: (1) Mobile personal computer equipment: This type of equipment belongs to the category of personal computers, with computing and processing functions, and generally also has the characteristics of mobile Internet access. Such terminals include: PDA, MID and UMPC equipment, etc., such as iPad; (2) other electronic devices with the function of generating images of real objects.

Those skilled in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, the program is stored in a storage medium, and includes several instructions to make a device ( It may be a single-chip microcomputer, a chip, etc.) or a processor (processor) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage media include: USB disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, etc., which can store program codes. medium.

In addition, the present invention also provides a non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the physical image generation method described in any one of the above.

In addition, the present invention also provides a computer program product, the computer program product includes a computing program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by the computer, the computer Execute the physical image generation method described in any one of the above.

Finally, it should be noted that those skilled in the art should understand that the embodiments of the present invention may be provided as methods, systems, or computer program products. Accordingly, the present invention can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may be embodied in one or more computer usable storage media having computer usable program code embodied therein. In the form of computer program products implemented on physical media (including but not limited to disk memory, optical memory, etc.).

For software implementation, the technology described in the embodiments of the present invention can be implemented through modules (such as procedures, functions, etc.) that execute the functions described in the embodiments of the present invention. Software codes may be stored in memory and executed by a processor. Memory can be implemented within the processor or external to the processor.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to the invention. It should be understood that each process and/or block in the flowchart and/or block diagram, and a combination of processes and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing equipment to produce a machine so that the instructions executed by the processor of the computer or other programmable data processing equipment Produce means for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer readable memory produce an article of manufacture including the instruction means , the instruction means implements the functions specified in one or more procedures of the flow chart and/or one or more blocks of the block diagram.

The above are only preferred embodiments of the present invention, and are not used to limit the implementation scope of the present invention. If the present invention is modified or equivalently replaced without departing from the spirit and scope of the present invention, it shall be covered by the protection of the patent scope of the present invention. in the range.

201-203: Steps

Claims (5)

A method for generating an object image, comprising: performing eigendecomposition on a first object image, obtaining a first reflection image and a first illumination image of the first object image; The first pixel value conversion rule is to perform M different transformations on the pixel values in the first illuminated image to obtain different M transformed second illuminated images; wherein, each of the M different transformations is transformed It is uniquely corresponding to one of the M transformed second irradiation images; M is a positive integer; according to the first reflection image and the M transformed second irradiation images, generate and M second real images that are different from the first real image; or, according to the N second pixel value transformation rules in the preset reflection transformation algorithm, make N times different pixel values in the first reflection image transformation to obtain different N transformed second reflection images; wherein, each transformation in the N different transformations is unique to one of the N transformed second reflection images Corresponding; N is a positive integer; according to the first illuminated image and the N transformed second reflection images, generate N second physical images different from the first real image; or, according to preset P third pixel value conversion rules in the illumination transformation algorithm, perform P different transformations on the pixel values in the first illumination image, and obtain different P transformed third illumination images; wherein, the P In different transformations, each transformation is uniquely corresponding to one of the P transformed third illumination images; P is a positive integer; according to the Q fourth pixel values in the preset reflection transformation algorithm Transformation rules, perform Q different transformations on the pixel values in the first reflected image, and obtain different Q transformed third reflected images image; wherein, each transformation in the Q times of different transformations is uniquely corresponding to a third illumination image in the Q transformed third illumination images; Q is a positive integer; according to the P transformed first Three irradiated images and the Q transformed third reflection images generate P*Q second physical images different from the first physical image; the second physical image is used as training data and input to An image recognition model; or, the M second physical images are used as training data and input to the image recognition model; or, the P*Q second physical images are used as training data and input to the image A recognition model; updating parameters of the image recognition model according to each second object image in the training data and the output result after the second object image is input to the image recognition model. An object image generation device, comprising: an acquisition module, configured to perform eigendecomposition on a first object image, and acquire a first reflection image and a first irradiation image of the first object image; a processing module, It is used to perform M different transformations on the pixel values in the first illumination image according to the M first pixel value transformation rules in the preset illumination transformation algorithm, and obtain different M transformed second illumination images; Wherein, each transformation in the M different transformations is uniquely corresponding to one of the M transformed second illumination images; M is a positive integer; according to the first reflected image and the M converted second illuminated images to generate M second real object images different from the first real object images; Perform N different transformations on the pixel values in a reflection image to obtain different N transformed second reflection images; wherein, each transformation in the N different transformations is the same as the N transformed second A second reflection image in the reflection image is uniquely corresponding; N is a positive integer; according to the first illumination image and The N transformed second reflection images generate N second real object images different from the first real object images; or, according to the P third pixel value transformation rules in the preset illumination transformation algorithm, for The pixel values in the first illuminated image are transformed differently P times to obtain different P transformed third illuminated images; wherein, each transformation in the P times different transformed is different from the P transformed transformed A third irradiation image in the third irradiation image is uniquely corresponding; P is a positive integer; according to the Q fourth pixel value conversion rule in the preset reflection transformation algorithm, the pixel value in the first reflection image is performed Q times of different transformations to obtain different Q transformed third reflected images; wherein, each transformation in the Q times of different transformations is related to one third irradiation in the Q transformed third irradiation images The image is uniquely corresponding; Q is a positive integer; according to the P transformed third irradiation images and the Q transformed third reflection images, P*Q first object images different from the first physical image are generated. Two physical images; it is also used to input the N second physical images as training data to the image recognition model; or, use the M second physical images as training data to input to the image recognition model ; Or, input the P*Q second physical images as training data into the image recognition model; according to each second physical image in the training data, input the second physical images into the The output result of the image recognition model is to update the parameters of the image recognition model. A physical image generating device, comprising: at least one processor; and a memory connected in communication with the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor At least one processor executes, so that the at least one processor can execute the first item of the patent scope of the application The method for generating the physical image. A non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the physical image generation method as described in item 1 of the scope of the patent application. A computer program product, the computer program product includes a computing program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer, the computer can execute The method for generating an image of a physical object as described in item 1.

Images